-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


Iljitsch van Beijnum wrote:
> On 18 mei 2009, at 23:17, Joe Touch wrote:
> 
>>> Not necessarily. PMTUD for tunnels is just an optimization. There is no
>>> reason why you couldn't take full size inner packets and encapsulate
>>> them in an outer packet that is then fragmented.
> 
>> You wouldn't know to do that unless you had a different idea of the MTU
>> at the endpoint. Or are you suggesting that ROHC do this (I don't think
>> it would)...
> 
> Suppose the MTU is 1500, IPsec takes 50 bytes and ROHC may add 4 bytes
> worst case but saves 32 bytes on average (completely made up numbers). Now:
> 
> - packets smaller than 1500 - 50 - 4 = 1446 bytes can always be carried
> without trouble, so nothing special happens
> 
> - packets larger than 1500 - 50 + 32 = 1482 bytes can never be carried
> without fragmentation, so send a too big message.
> 
> - packets between 1446 and 1482 bytes will usually compress to something
> that can be carried without fragmentation, so we don't send a too big
> message. However, they may not compress, in which case the resulting
> packet is larger than 1500 bytes. We then fragment the packet upon
> sending and reassemble before detunneling.
> 
> The whole thing can be made dynamic so in the range 1446 - 1482 we don't
> always fragment but rather when we see that we fragment too much, we
> lower the MTU value reported back to sources until the level of
> fragmentation is within reason.

That's fine if that's what ROHC already does.

It's not fine if this is what you expect and ROHC dumps the too-big
packets between 1446 and 1482 on the floor.

I agree that some fragmentation like this might actually be a big win.
However, we're also assuming that the receiver reassembly happens fast
enough that the packets aren't delayed - i.e., if they are sent 'slow
path' in the receiver (if there is such a thing, and there could be,
e.g., on hardware-assisted hosts), then this would muck up the ROHC
mechanism...

Joe
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkoR1SYACgkQE5f5cImnZrutOACg4u3dAoVUCBdYqzZyZPy+UDMn
5PQAoNyAcTYdF6vfg7Xo6mtYRfh+JuQj
=CAik
-----END PGP SIGNATURE-----

Reply via email to